Thought Leaders in Big Data: Interview with Brian Bulkowski and Srini Srinivasan, Co-founders of Aerospike (Part 3)

Sramana Mitra: How is this all handled? You said Aerospike cluster, does that mean you deliver your service on the cloud?

Srini Srinivasan: We build software and we license our software to our customers. They put it on their hardware and run it. However, the software is self-managing. It is very easy to set up, and it is extremely efficient to manage operationally. We have about three dozen active mission-critical deployments worldwide across 50 to 60 data centers and across many clusters. All of that is running Aerospike day in and day out. Not much operational overhead exists compared to other comparable products because we have “baked in” a lot of the service ability or basic algorithms of how you run a system with no downtime. We only do rolling upgrades. There is no such thing as a maintenance window.

SM: And how do you charge clients?

Brian Bulkowski: We charge them by the data and volume. When we are working out the pricing – what would be best for this product – what we find is that if you price by number of cores or machines, which is common in distributed databases, the problem is that operations people will try to use as few machines as possible. Then you start getting the operational decisions confused with the licensing decisions. What we found is that you really need to let the operation’s staff make decisions for their own reasons and normal patterns without thinking about what the licenses are. When you charge by data, you are basically charging by the business usefulness of the project.

SM: And you have the auditing of all that built into the product?

SS: We are of course able to measure the data our system is serving. We don’t actually have any code in there to disable it.

SM: But you get a report?

SS: We can get a report. But typically we don’t enforce it.

SM: But you need to be able to bill based on how much data has been used.

BB: Oracle, for example, does it the same way. We have a relationship with many of these big companies. They are trustworthy companies, and many of them are public. They don’t go around evading license fees.

SM: So, they estimate how much they are going to use and pay accordingly?

SS: It is what they buy as a capacity, and they may or may not use that much. We have the right to audit it once per year, but we never have to do that because most of our operations team has direct access to those machines anyways.

SM: So it is not exactly utilization based. It is kind of an estimate of much they use.

BB: We found that trusting our customers the way they trust us is good business.

SM: I am just trying to understand your model.

SS: Somebody could buy a one-terabyte license, but they could only be using half a terabyte. We don’t measure that.

SM: It is like buying a cellphone plan. We buy a certain plan, and whether we use it or not we pay.

SS: That is the model.

SM: It is not like I pay for 20, 30 or 40 messages.

SS: It is not at that level. It is more at the contract level.

SM: Are there any other customers you would like to talk about?

SS: Of course. The interesting thing is that our contracts are based on the amount of data. People can choose different kinds of machines to put in their cluster. AppNexus, for example, is using a heavily Flash-based system, whereas The Trade Desk uses DRAM-based systems to run their clusters. They also run large volumes.

This segment is part 3 in the series : Thought Leaders in Big Data: Interview with Brian Bulkowski and Srini Srinivasan, Co-founders of Aerospike
1 2 3 4 5 6 7